From 2e22502c080f27afeab5e6f11e618fb7bc7aea53 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Mon, 23 Nov 2015 19:32:51 +0530 Subject: ARC: dw2 unwind: Remove falllback linear search thru FDE entries Fixes STAR 9000953410: "perf callgraph profiling causing RCU stalls" | perf record -g -c 15000 -e cycles /sbin/hackbench | | INFO: rcu_preempt self-detected stall on CPU | 1: (1 GPs behind) idle=609/140000000000002/0 softirq=2914/2915 fqs=603 | Task dump for CPU 1: in-kernel dwarf unwinder has a fast binary lookup and a fallback linear search (which iterates thru each of ~11K entries) thus takes 2 orders of magnitude longer (~3 million cycles vs. 2000). Routines written in hand assembler lack dwarf info (as we don't support assembler CFI pseudo-ops yet) fail the unwinder binary lookup, hit linear search, failing nevertheless in the end. However the linear search is pointless as binary lookup tables are created from it in first place. It is impossible to have binary lookup fail while succeed the linear search. It is pure waste of cycles thus removed by this patch. This manifested as RCU stalls / NMI watchdog splat when running hackbench under perf with callgraph profiling. The triggering condition was perf counter overflowing in routine lacking dwarf info (like memset) leading to patheic 3 million cycle unwinder slow path and by the time it returned new interrupts were already pending (Timer, IPI) and taken rightaway. The original memset didn't make forward progress, system kept accruing more interrupts and more unwinder delayes in a vicious feedback loop, ultimately triggering the NMI diagnostic. Cc: stable@vger.kernel.org Signed-off-by: Vineet Gupta --- arch/arc/kernel/unwind.c | 37 ++++--------------------------------- 1 file changed, 4 insertions(+), 33 deletions(-) (limited to 'arch/arc/kernel/unwind.c') diff --git a/arch/arc/kernel/unwind.c b/arch/arc/kernel/unwind.c index 93c6ea52b671..7352475451f6 100644 --- a/arch/arc/kernel/unwind.c +++ b/arch/arc/kernel/unwind.c @@ -986,42 +986,13 @@ int arc_unwind(struct unwind_frame_info *frame) (const u8 *)(fde + 1) + *fde, ptrType); - if (pc >= endLoc) + if (pc >= endLoc) { fde = NULL; - } else - fde = NULL; - } - if (fde == NULL) { - for (fde = table->address, tableSize = table->size; - cie = NULL, tableSize > sizeof(*fde) - && tableSize - sizeof(*fde) >= *fde; - tableSize -= sizeof(*fde) + *fde, - fde += 1 + *fde / sizeof(*fde)) { - cie = cie_for_fde(fde, table); - if (cie == &bad_cie) { cie = NULL; - break; } - if (cie == NULL - || cie == ¬_fde - || (ptrType = fde_pointer_type(cie)) < 0) - continue; - ptr = (const u8 *)(fde + 2); - startLoc = read_pointer(&ptr, - (const u8 *)(fde + 1) + - *fde, ptrType); - if (!startLoc) - continue; - if (!(ptrType & DW_EH_PE_indirect)) - ptrType &= - DW_EH_PE_FORM | DW_EH_PE_signed; - endLoc = - startLoc + read_pointer(&ptr, - (const u8 *)(fde + - 1) + - *fde, ptrType); - if (pc >= startLoc && pc < endLoc) - break; + } else { + fde = NULL; + cie = NULL; } } } -- cgit v1.2.3 From bc79c9a7216562a2035d2f64f73626613c1300d0 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Fri, 11 Dec 2015 18:04:22 +0530 Subject: ARC: dw2 unwind: Reinstante unwinding out of modules The fix which removed linear searching of dwarf (because binary lookup data always exists) missed out on the fact that modules don't get the binary lookup tables info. This caused unwinding out of modules to stop working. So add binary lookup header setup (equivalent of eh_frame_hdr setup) to modules as well. While at it, confine the header setup to within unwinder code, reducing one API exposed out of unwinder code. Fixes: 2e22502c080f ARC: dw2 unwind: Remove falllback linear search thru FDE entries Cc: Signed-off-by: Vineet Gupta --- arch/arc/kernel/unwind.c | 40 ++++++++++++++++++++++++++-------------- 1 file changed, 26 insertions(+), 14 deletions(-) (limited to 'arch/arc/kernel/unwind.c') diff --git a/arch/arc/kernel/unwind.c b/arch/arc/kernel/unwind.c index 7352475451f6..9f9ecc15556e 100644 --- a/arch/arc/kernel/unwind.c +++ b/arch/arc/kernel/unwind.c @@ -170,6 +170,23 @@ static struct unwind_table *find_table(unsigned long pc) static unsigned long read_pointer(const u8 **pLoc, const void *end, signed ptrType); +static void init_unwind_hdr(struct unwind_table *table, + void *(*alloc) (unsigned long)); + +/* + * wrappers for header alloc (vs. calling one vs. other at call site) + * to elide section mismatches warnings + */ +static void *__init unw_hdr_alloc_early(unsigned long sz) +{ + return __alloc_bootmem_nopanic(sz, sizeof(unsigned int), + MAX_DMA_ADDRESS); +} + +static void *unw_hdr_alloc(unsigned long sz) +{ + return kmalloc(sz, GFP_KERNEL); +} static void init_unwind_table(struct unwind_table *table, const char *name, const void *core_start, unsigned long core_size, @@ -209,6 +226,8 @@ void __init arc_unwind_init(void) __start_unwind, __end_unwind - __start_unwind, NULL, 0); /*__start_unwind_hdr, __end_unwind_hdr - __start_unwind_hdr);*/ + + init_unwind_hdr(&root_table, unw_hdr_alloc_early); } static const u32 bad_cie, not_fde; @@ -241,8 +260,8 @@ static void swap_eh_frame_hdr_table_entries(void *p1, void *p2, int size) e2->fde = v; } -static void __init setup_unwind_table(struct unwind_table *table, - void *(*alloc) (unsigned long)) +static void init_unwind_hdr(struct unwind_table *table, + void *(*alloc) (unsigned long)) { const u8 *ptr; unsigned long tableSize = table->size, hdrSize; @@ -300,9 +319,11 @@ static void __init setup_unwind_table(struct unwind_table *table, hdrSize = 4 + sizeof(unsigned long) + sizeof(unsigned int) + 2 * n * sizeof(unsigned long); + header = alloc(hdrSize); if (!header) return; + header->version = 1; header->eh_frame_ptr_enc = DW_EH_PE_abs | DW_EH_PE_native; header->fde_count_enc = DW_EH_PE_abs | DW_EH_PE_data4; @@ -342,18 +363,6 @@ static void __init setup_unwind_table(struct unwind_table *table, table->header = (const void *)header; } -static void *__init balloc(unsigned long sz) -{ - return __alloc_bootmem_nopanic(sz, - sizeof(unsigned int), - __pa(MAX_DMA_ADDRESS)); -} - -void __init arc_unwind_setup(void) -{ - setup_unwind_table(&root_table, balloc); -} - #ifdef CONFIG_MODULES static struct unwind_table *last_table; @@ -377,6 +386,8 @@ void *unwind_add_table(struct module *module, const void *table_start, table_start, table_size, NULL, 0); + init_unwind_hdr(table, unw_hdr_alloc); + #ifdef UNWIND_DEBUG unw_debug("Table added for [%s] %lx %lx\n", module->name, table->core.pc, table->core.range); @@ -439,6 +450,7 @@ void unwind_remove_table(void *handle, int init_only) info.init_only = init_only; unlink_table(&info); /* XXX: SMP */ + kfree(table->header); kfree(table); } -- cgit v1.2.3 From 323f41f9e7d0cb5b1d1586aded6682855f1e646d Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Wed, 16 Dec 2015 17:17:00 +0530 Subject: ARC: dw2 unwind: Ignore CIE version !=1 gracefully instead of bailing ARC dwarf unwinder only supports CIE version == 1 The boot time dwarf sanitizer (part of binary lookup table constructor) would simply bail if it saw CIE version == 3, rendering unwinder with a NULL lookup table. It seems libgcc linked with kernel does have such entries. With fallback linear search removed, and a NULL binary lookup table, unwinder fails to generate any stack trace. So allow graceful ignoring of unsupported CIE entries. This problem was initially seen in Alexey's setup (and not mine) as he was using buildroot built toolchain (libgcc) which doesn't get built with CFLAGS_FOR_TARGET="-gdwarf-2 which is my default Fixes STAR 9000985048: "kernel unwinder broken with stock tools" Fixes: 2e22502c080f ARC: dw2 unwind: Remove falllback linear search thru FDE entries Reported-by Alexey Brodkin Cc: Signed-off-by: Vineet Gupta --- arch/arc/kernel/unwind.c | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-) (limited to 'arch/arc/kernel/unwind.c') diff --git a/arch/arc/kernel/unwind.c b/arch/arc/kernel/unwind.c index 9f9ecc15556e..cf2828ab0905 100644 --- a/arch/arc/kernel/unwind.c +++ b/arch/arc/kernel/unwind.c @@ -293,13 +293,13 @@ static void init_unwind_hdr(struct unwind_table *table, const u32 *cie = cie_for_fde(fde, table); signed ptrType; - if (cie == ¬_fde) + if (cie == ¬_fde) /* only process FDE here */ continue; if (cie == NULL || cie == &bad_cie) - return; + continue; /* say FDE->CIE.version != 1 */ ptrType = fde_pointer_type(cie); if (ptrType < 0) - return; + continue; ptr = (const u8 *)(fde + 2); if (!read_pointer(&ptr, (const u8 *)(fde + 1) + *fde, @@ -343,6 +343,10 @@ static void init_unwind_hdr(struct unwind_table *table, if (fde[1] == 0xffffffff) continue; /* this is a CIE */ + + if (*(u8 *)(cie + 2) != 1) + continue; /* FDE->CIE.version not supported */ + ptr = (const u8 *)(fde + 2); header->table[n].start = read_pointer(&ptr, (const u8 *)(fde + 1) + @@ -519,7 +523,8 @@ static const u32 *cie_for_fde(const u32 *fde, const struct unwind_table *table) if (*cie <= sizeof(*cie) + 4 || *cie >= fde[1] - sizeof(*fde) || (*cie & (sizeof(*cie) - 1)) - || (cie[1] != 0xffffffff)) + || (cie[1] != 0xffffffff) + || ( *(u8 *)(cie + 2) != 1)) /* version 1 supported */ return NULL; /* this is not a (valid) CIE */ return cie; } -- cgit v1.2.3 From 2d64affc92d6281d2eca4faad326b981773bf040 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Mon, 21 Dec 2015 13:29:44 +0530 Subject: Revert "ARC: dw2 unwind: Ignore CIE version !=1 gracefully instead of bailing" Blingly ignoring CIE.version != 1 was a bad idea. It still leaves "desirability" when running perf with callgraphing where libgcc symbols might show in hotspot. More importantly, basic CIE.version == 3 support already exists in code: | | retAddrReg = state.version <= 1 ? *ptr++ : get_uleb128(&ptr, end); | Next commit with simply add continue-not-bail for CIE.version != 1 This reverts commit 323f41f9e7d0cb5b1d1586aded6682855f1e646d. --- arch/arc/kernel/unwind.c | 13 ++++--------- 1 file changed, 4 insertions(+), 9 deletions(-) (limited to 'arch/arc/kernel/unwind.c') diff --git a/arch/arc/kernel/unwind.c b/arch/arc/kernel/unwind.c index cf2828ab0905..9f9ecc15556e 100644 --- a/arch/arc/kernel/unwind.c +++ b/arch/arc/kernel/unwind.c @@ -293,13 +293,13 @@ static void init_unwind_hdr(struct unwind_table *table, const u32 *cie = cie_for_fde(fde, table); signed ptrType; - if (cie == ¬_fde) /* only process FDE here */ + if (cie == ¬_fde) continue; if (cie == NULL || cie == &bad_cie) - continue; /* say FDE->CIE.version != 1 */ + return; ptrType = fde_pointer_type(cie); if (ptrType < 0) - continue; + return; ptr = (const u8 *)(fde + 2); if (!read_pointer(&ptr, (const u8 *)(fde + 1) + *fde, @@ -343,10 +343,6 @@ static void init_unwind_hdr(struct unwind_table *table, if (fde[1] == 0xffffffff) continue; /* this is a CIE */ - - if (*(u8 *)(cie + 2) != 1) - continue; /* FDE->CIE.version not supported */ - ptr = (const u8 *)(fde + 2); header->table[n].start = read_pointer(&ptr, (const u8 *)(fde + 1) + @@ -523,8 +519,7 @@ static const u32 *cie_for_fde(const u32 *fde, const struct unwind_table *table) if (*cie <= sizeof(*cie) + 4 || *cie >= fde[1] - sizeof(*fde) || (*cie & (sizeof(*cie) - 1)) - || (cie[1] != 0xffffffff) - || ( *(u8 *)(cie + 2) != 1)) /* version 1 supported */ + || (cie[1] != 0xffffffff)) return NULL; /* this is not a (valid) CIE */ return cie; } -- cgit v1.2.3 From 6d0d506012c93d3393b8d3d0cac62e46e541c5b6 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Mon, 21 Dec 2015 13:40:05 +0530 Subject: ARC: dw2 unwind: Don't bail for CIE.version != 1 The rudimentary CIE.version == 3 handling is already present in code (for return address register specification) Signed-off-by: Vineet Gupta --- arch/arc/kernel/unwind.c | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-) (limited to 'arch/arc/kernel/unwind.c') diff --git a/arch/arc/kernel/unwind.c b/arch/arc/kernel/unwind.c index 9f9ecc15556e..f34599abe182 100644 --- a/arch/arc/kernel/unwind.c +++ b/arch/arc/kernel/unwind.c @@ -600,9 +600,6 @@ static signed fde_pointer_type(const u32 *cie) const u8 *ptr = (const u8 *)(cie + 2); unsigned version = *ptr; - if (version != 1) - return -1; /* unsupported */ - if (*++ptr) { const char *aug; const u8 *end = (const u8 *)(cie + 1) + *cie; @@ -1014,9 +1011,7 @@ int arc_unwind(struct unwind_frame_info *frame) ptr = (const u8 *)(cie + 2); end = (const u8 *)(cie + 1) + *cie; frame->call_frame = 1; - if ((state.version = *ptr) != 1) - cie = NULL; /* unsupported version */ - else if (*++ptr) { + if (*++ptr) { /* check if augmentation size is first (thus present) */ if (*ptr == 'z') { while (++ptr < end && *ptr) { -- cgit v1.2.3 From 6b538db7c6b091960c57b23feae5431fc82286e7 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Mon, 21 Dec 2015 13:52:01 +0530 Subject: ARC: dw2 unwind: Catch Dwarf SNAFUs early Instead of seeing empty stack traces, let kernel fail early so dwarf issues can be fixed sooner Signed-off-by: Vineet Gupta --- arch/arc/kernel/unwind.c | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-) (limited to 'arch/arc/kernel/unwind.c') diff --git a/arch/arc/kernel/unwind.c b/arch/arc/kernel/unwind.c index f34599abe182..5eb707640e9c 100644 --- a/arch/arc/kernel/unwind.c +++ b/arch/arc/kernel/unwind.c @@ -296,10 +296,10 @@ static void init_unwind_hdr(struct unwind_table *table, if (cie == ¬_fde) continue; if (cie == NULL || cie == &bad_cie) - return; + goto ret_err; ptrType = fde_pointer_type(cie); if (ptrType < 0) - return; + goto ret_err; ptr = (const u8 *)(fde + 2); if (!read_pointer(&ptr, (const u8 *)(fde + 1) + *fde, @@ -315,14 +315,14 @@ static void init_unwind_hdr(struct unwind_table *table, } if (tableSize || !n) - return; + goto ret_err; hdrSize = 4 + sizeof(unsigned long) + sizeof(unsigned int) + 2 * n * sizeof(unsigned long); header = alloc(hdrSize); if (!header) - return; + goto ret_err; header->version = 1; header->eh_frame_ptr_enc = DW_EH_PE_abs | DW_EH_PE_native; @@ -361,6 +361,10 @@ static void init_unwind_hdr(struct unwind_table *table, table->hdrsz = hdrSize; smp_wmb(); table->header = (const void *)header; + return; + +ret_err: + panic("Attention !!! Dwarf FDE parsing errors\n");; } #ifdef CONFIG_MODULES -- cgit v1.2.3